近年来，随着食品安全得到越来越多大众的重视，促进了溯源技术在食品行业中快速发展。为研究鳙鱼自动溯源算法，本文选择以鳙鱼图像为研究对象以探索实现自动鳙鱼溯源的算法，并分析了鳙鱼图像数据的特点，提出两点亟待解决的问题：网络模型选择、数据迁移等。故本文针对目前存在问题提出了一种基于深度学习的自动鳙鱼溯源算法，并在实验测试中算法模型准确率可达到96.865%。在本文中，针对模型选择问题，开展研究了国内外主流的深度学习框架网络，通过对比分析选用DenseNet网络，通过DenseNet网络模型可以将卷积神经网络每层学习到的特征值向后传递，并且即能在长期学习过程中保持学习最初有效的特征，提高特征提取的效率，还能大大减少了学习成本。针对鳙鱼图像数据迁移问题，本文采用了迁移学习的思想，将训练模型分解为在线训练和离线训练两个步骤，不断学习输入来的鳙鱼特征，以保证在不同营养型的水域，仍然能达到很高的准确率。

In recent years, with the public paying more and more attention to food safety, the traceability technology has promoted the rapid development in the food industry.In order to study the automatic traceability algorithm of bighead fish, this paper chose the image of bighead fish as the research object to explore the algorithm of automatic traceability of bighead fish, analyzed the characteristics of image data of bighead fish, and proposed two problems to be solved urgently: network model selection, data migration, etc.Therefore, this paper proposes an automatic bighead carp traceability algorithm based on deep learning to solve the existing problems, and the accuracy of the algorithm model can reach 96.865% in the experimental test.In this paper, aiming at the problem of model selection, the mainstream deep learning framework networks at home and abroad are studied. By comparative analysis, the Densenet network is selected. Through the Densenet network model, the eigenvalues learned at each layer of the convolutional neural network can be transmitted back, and the initial effective features of learning can be maintained in the long-term learning process.The efficiency of feature extraction can be improved, and the learning cost can be greatly reduced.To solve the problem of image data migration of bighead fish, this paper adopts the idea of transfer learning and decompositions the training model into two steps: online training and offline training, so as to continuously learn the input characteristics of bighead fish, so as to ensure that high accuracy can still be achieved in the waters with different nutritional types.

重庆市技术创新与应用发展专项重点项目（cstc2019jscx-dxwtBX0034）